US20060136209A1 - Methodology for generating enhanced demiphone acoustic models for speech recognition - Google Patents
Methodology for generating enhanced demiphone acoustic models for speech recognition Download PDFInfo
- Publication number
- US20060136209A1 US20060136209A1 US11/013,888 US1388804A US2006136209A1 US 20060136209 A1 US20060136209 A1 US 20060136209A1 US 1388804 A US1388804 A US 1388804A US 2006136209 A1 US2006136209 A1 US 2006136209A1
- Authority
- US
- United States
- Prior art keywords
- demiphone
- succeeding
- acoustic
- dominant
- models
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/14—Speech classification or search using statistical models, e.g. Hidden Markov Models [HMMs]
- G10L15/142—Hidden Markov Models [HMMs]
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
- G10L2015/022—Demisyllables, biphones or triphones being the recognition units
Definitions
- This invention relates generally to electronic speech recognition systems, and relates more particularly to a methodology for generating enhanced demiphone acoustic models for speech recognition.
- Voice-controlled operation of electronic devices often provides a desirable interface for system users to control and interact with electronic devices.
- voice-controlled operation of an electronic device may allow a user to perform other tasks simultaneously, or can be advantageous in certain types of operating environments.
- hands-free operation of electronic devices may also be desirable for users who have physical limitations or other special requirements.
- Hands-free operation of electronic devices may be implemented by various speech-activated electronic devices.
- Speech-activated electronic devices advantageously allow users to interface with electronic devices in situations where it would be inconvenient or potentially hazardous to utilize a traditional input device.
- effectively implementing such speech recognition systems creates substantial challenges for system designers.
- a methodology for generating enhanced demiphone acoustic models for speech recognition.
- a phone set of specific phones are selectively defined for utilization by a speech recognition engine to perform various speech recognition procedures.
- a representative training database is created or provided by utilizing any effective techniques. Then, an acoustic model generator or other appropriate entity analyzes the training database to generate a separate decision tree for each state of acoustic models that represent the defined phone set.
- the acoustic model generator analyzes the decision trees to define appropriate dominance configurations for corresponding enhanced demiphone acoustic models.
- the foregoing configurations include a preceding-dominant enhanced demiphone configuration and a succeeding-dominant enhanced demiphone configuration.
- the acoustic model generator then implements the enhanced demiphone acoustic models using the dominance configurations previously defined.
- the succeeding-dominant (right-dominant) enhanced demiphone includes a preceding (left) demiphone with relatively fewer states (e.g. one or zero states) and a succeeding (right) demiphone with relatively more states (e.g. two or three states).
- the preceding-dominant (left-dominant) enhanced demiphone includes a preceding (left) demiphone with relatively more states (e.g. two or three states) and a succeeding (right) demiphone with relatively fewer states (zero or one state).
- the speech recognition engine may then utilize the enhanced demiphone acoustic models to accurately and efficiently perform speech recognition procedures. For at least the foregoing reasons, the present invention therefore provides an improved system and method for generating enhanced demiphone acoustic models for speech recognition.
- FIG. 1 is a block diagram for one embodiment of an electronic device, in accordance with the present invention.
- FIG. 2 is a block diagram for one embodiment of the memory of FIG. 1 , in accordance with the present invention.
- FIG. 3 is a block diagram for one embodiment of the speech recognition engine of FIG. 2 , in accordance with the present invention.
- FIG. 4 is a block diagram illustrating functionality of the speech recognition engine of FIG. 3 , in accordance with one embodiment of the present invention
- FIG. 5 is a diagram for one embodiment of a triphone acoustic model
- FIG. 6 is a diagram for one embodiment of a conventional demiphone acoustic model
- FIG. 7 is a diagram for one embodiment of a right-dominant enhanced demiphone acoustic model, in accordance with the present invention.
- FIG. 8 is a diagram for one embodiment of a left-dominant enhanced demiphone acoustic model, in accordance with the present invention.
- FIG. 9 is a diagram for one embodiment of a decision tree, in accordance with the present invention.
- FIG. 10 is a flowchart of method steps for generating enhanced demiphone acoustic models for speech recognition, in accordance with one embodiment of the present invention.
- the present invention relates to an improvement in speech recognition systems.
- the following description is presented to enable one of ordinary skill in the art to make and use the invention, and is provided in the context of a patent application and its requirements.
- Various modifications to the embodiments disclosed herein will be apparent to those skilled in the art, and the generic principles herein may be applied to other embodiments.
- the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features described herein.
- the present invention comprises a system and method for effectively performing speech recognition procedures, and includes enhanced demiphone acoustic models that a speech recognition engine utilizes to perform the speech recognition procedures.
- the enhanced demiphone acoustic models each have three states that are collectively arranged to form a left demiphone and a right demiphone.
- An acoustic model generator utilizes a decision tree for analyzing speech context information from a training database. To accurately model speech characteristics, the acoustic model generator then configures each of the enhanced demiphone acoustic models as either a right-dominant enhanced demiphone acoustic model or a left-dominant enhanced demiphone acoustic model.
- FIG. 1 a block diagram for one embodiment of an electronic device 110 is shown, according to the present invention.
- the FIG. 1 embodiment includes, but is not limited to, a sound sensor 112 , a control module 114 , and a display 134 .
- electronic device 110 may readily include various other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with the FIG. 1 embodiment.
- electronic device 110 may be embodied as any appropriate electronic device or system.
- electronic device 110 may be implemented as a computer device, a personal digital assistant (PDA), a cellular telephone, a television, a game console, and as part of entertainment robots such as AIBOTM and QRIOTM by Sony Corporation.
- PDA personal digital assistant
- AIBOTM and QRIOTM part of entertainment robots
- electronic device 110 utilizes sound sensor 112 to detect and convert ambient sound energy into corresponding audio data.
- the captured audio data is then transferred over system bus 124 to CPU 122 , which responsively performs various processes and functions with the captured audio data, in accordance with the present invention.
- control module 114 includes, but is not limited to, a central processing unit (CPU) 122 (electronic data processor), a memory 130 , and one or more input/output interface(s) (I/O) 126 .
- Display 134 , CPU 122 , memory 130 , and I/O 126 are each coupled to, and communicate, via common system bus 124 .
- control module 114 may readily include various other components in addition to, or instead of, those components discussed in conjunction with the FIG. 1 embodiment.
- CPU 122 is implemented to include any appropriate microprocessor device. Alternately, CPU 122 may be implemented using any other appropriate technology. For example, CPU 122 may be implemented as an application-specific integrated circuit (ASIC) or other appropriate electronic device.
- ASIC application-specific integrated circuit
- I/O 126 provides one or more effective interfaces for facilitating bi-directional communications between electronic device 110 and any external entity, including a system user or another electronic device. I/O 126 may be implemented using any appropriate input and/or output devices. The functionality and utilization of electronic device 110 are further discussed below in conjunction with FIGS. 2-10 .
- Memory 130 may comprise any desired storage-device configurations, including, but not limited to, random access memory (RAM), read-only memory (ROM), and storage devices such as floppy discs or hard disc drives, and SONY® MEMORY STICK®.
- RAM random access memory
- ROM read-only memory
- AM acoustic model
- FIG. 2 embodiment memory 130 stores a device application 210 , a speech recognition engine 214 , an acoustic model (AM) generator 222 , and a training database 226 .
- AM acoustic model
- memory 130 may readily store other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with the FIG. 2 embodiment.
- device application 210 includes program instructions that are executed by CPU 122 ( FIG. 1 ) to perform various functions and operations for electronic device 110 .
- the particular nature and functionality of device application 210 typically varies depending upon factors such as the type and particular use of the corresponding electronic device 110 .
- speech recognition engine 214 includes one or more software modules that are executed by CPU 122 to analyze and recognize input sound data. Certain embodiments of speech recognition engine 214 are further discussed below in conjunction with FIGS. 3-4 .
- electronic device 110 may utilize AM generator 222 to generate enhanced demiphone acoustic models for use by speech recognition engine 214 in performing speech recognition procedures.
- training database 226 includes an extensive selection of representative recorded speech samples. The generation of enhanced demiphone acoustic models by AM generator 222 is further discussed below in conjunction with FIGS. 7-10 .
- Speech recognition engine 214 includes, but is not limited to, a feature extractor 310 , an endpoint detector 312 , a recognizer 314 , acoustic models 336 , dictionary 340 , and language models 344 .
- speech recognition engine 214 may readily include various other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with the FIG. 3 embodiment.
- sound sensor 112 ( FIG. 1 ) provides digital speech data to feature extractor 310 via system bus 124 .
- Feature extractor 310 responsively generates corresponding representative feature vectors, which are provided to recognizer 314 via path 320 .
- Feature extractor 310 further provides the speech data to endpoint detector 312 , and endpoint detector 312 responsively identifies endpoints of utterances represented by the speech data to indicate the beginning and end of an utterance in time. Endpoint detector 312 then provides the endpoints to recognizer 314 .
- recognizer 314 is configured to recognize words in a vocabulary that is represented in dictionary 340 .
- the vocabulary represented in dictionary 340 corresponds to any desired sentences, word sequences, commands, instructions, narration, or other audible sounds that are supported for speech recognition by speech recognition engine 214 .
- each word from dictionary 340 is associated with a corresponding phone string (string of individual phones) which represents the pronunciation of that word.
- Acoustic models 336 (such as Hidden Markov Models) for each of the phones are selected and combined to create the foregoing phone strings for accurately representing pronunciations of words in dictionary 340 .
- acoustic models 336 are implemented as enhanced demiphone acoustic models. Recognizer 314 compares input feature vectors from line 320 with the entries (phone strings) from dictionary 340 to determine which word produces the highest recognition score. The word corresponding to the highest recognition score may thus be identified as the recognized word.
- Speech recognition engine 214 also utilizes language models 344 as a recognition grammar to determine specific recognized word sequences that are supported by speech recognition engine 214 .
- the recognized sequences of vocabulary words may then be output as recognition results from recognizer 314 via path 332 .
- the operation and implementation of recognizer 314 , dictionary 340 , and language models 344 are further discussed below in conjunction with FIG. 4 .
- FIG. 4 a block diagram illustrating functionality of the FIG. 3 speech recognition engine 214 is shown, in accordance with one embodiment of the present invention.
- the present invention may readily perform speech recognition procedures using various techniques or functionalities in addition to, or instead of, certain techniques or functionalities discussed in conjunction with the FIG. 4 embodiment.
- speech recognition engine 214 receives speech data from sound sensor 112 , as discussed above in conjunction with FIG. 3 .
- Recognizer 314 FIG. 3
- speech recognition engine 214 sequentially compares segments of the input speech data with acoustic models 336 to identify a series of phones (phone strings) that represent the input speech data.
- acoustic models 336 are implemented as enhanced demiphone acoustic models, as discussed below in conjunction with FIGS. 7-10 .
- Recognizer 314 references dictionary 340 to look up recognized vocabulary words that correspond to the identified phone strings.
- the recognizer 314 then utilizes language models 344 as a recognition grammar to form the recognized vocabulary words into word sequences, such as sentences, phrases, commands, or narration, which are supported by speech recognition engine 214 .
- language models 344 as a recognition grammar to form the recognized vocabulary words into word sequences, such as sentences, phrases, commands, or narration, which are supported by speech recognition engine 214 .
- Various techniques for effectively generating enhanced demiphone acoustic models are further discussed below in conjunction with FIGS. 7-10 .
- triphone AM 512 represents a given phone from a supported phone set that is used to implement a speech recognition engine.
- Triphone AM 512 includes a first state 516 ( a ), a second state 516 ( b ), and a third state 516 ( c ) that collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in the FIG. 5 embodiment.
- Each state 516 of triphone AM 512 is defined with respect to a phone context that includes information from both a preceding phone and a succeeding phone. In other words, all three states 516 of the current phone that is modeled by triphone AM 512 are based upon context information from both an immediately preceding phone and an immediately succeeding phone.
- conventional demiphone AM 612 represents a given phone from a supported phone set that is used to implement a speech recognition engine.
- Conventional demiphone AM 612 includes a conventional left demiphone 620 ( a ) and a conventional right demiphone 620 ( b ).
- the conventional left demiphone 620 ( a ) in turn includes a first state 616 ( a ) and a second state 616 ( b ), while conventional right demiphone 620 ( b ) includes a third state 616 ( c ) and a fourth state 616 ( d ).
- First state 616 ( a ), second state 616 ( b ), third state 616 ( c ), and fourth state 616 ( d ) collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in the FIG. 6 embodiment.
- Each state 616 ( a ) and 616 ( b ) of conventional left demiphone AM 620 ( a ) is defined with respect to a phone context that includes information only from a preceding phone.
- each state 616 ( c ) and 616 ( d ) of conventional right demiphone AM 620 ( b ) is defined with respect to a phone context that includes information from only a succeeding phone.
- left demiphone 620 ( a ) of conventional demiphone AM 612 is based only upon context information from an immediately preceding phone with respect to the current phone that is modeled by conventional demiphone AM 612
- right demiphone 620 ( b ) of conventional demiphone AM 612 is based only upon context information from an immediately succeeding phone with respect to the current phone that is modeled by conventional demiphone AM 612 .
- conventional demiphone AMs 612 utilize the same number of states 616 to implement both conventional left demiphone 620 ( a ) and conventional right demiphone 620 ( b ), an equal level of dominance is accorded to both the preceding phone and the succeeding phone. In actuality, during most speech contexts, the foregoing equal level of dominance does not accurately reflect true acoustic characteristics of speech. Therefore, conventional demiphone AMs 612 do not optimally model speech, and speech recognition procedures may exhibit reduced accuracy as a result. To address this problem, improved techniques for implementing enhanced demiphone acoustic models are discussed below in conjunction with FIGS. 7-10 .
- FIG. 7 a diagram for one embodiment of a right-dominant enhanced demiphone acoustic model (AM) 712 ( a ) is shown, in accordance with one embodiment of the present invention.
- AMs 712 ( a ) may be implemented with configurations, elements, or functionalities in addition to, or instead of, certain configurations, elements, or functionalities discussed in conjunction with the FIG. 7 embodiment.
- right-dominant enhanced demiphone AM 712 ( a ) represents a given phone from a supported phone set that is used to implement acoustic models 336 and dictionary 340 from speech recognition engine 214 , as discussed above in conjunction with FIGS. 3 and 4 .
- Right-dominant enhanced demiphone AM 712 ( a ) includes a left demiphone 720 ( a ) and a right demiphone 720 ( b ).
- the left demiphone 720 ( a ) in turn includes only a first state 716 ( a ), while right demiphone 720 ( b ) includes a second state 716 ( b ) and a third state 716 ( c ).
- First state 716 ( a ), second state 716 ( b ), and third state 716 ( c ) collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in the FIG. 7 embodiment.
- State 716 ( a ) of left demiphone AM 720 ( a ) is defined with respect to a phone context that includes information only from a preceding phone.
- each state 716 ( b ) and 716 ( c ) of right demiphone AM 720 ( b ) is defined with respect to a phone context that includes information from only a succeeding phone.
- left demiphone 720 ( a ) of right-dominant enhanced demiphone AM 712 ( a ) is based upon context information from only an immediately preceding phone with respect to the current phone that is modeled by right-dominant enhanced demiphone AM 712 ( a )
- right demiphone 720 ( b ) of right-dominant enhanced demiphone AM 712 ( a ) is based upon context information from only an immediately succeeding phone with respect to the current phone that is modeled by right-dominant enhanced demiphone AM 712 ( a ).
- right-dominant demiphone AMs 712 ( a ) utilize a different number of states 716 to implement left demiphone 720 ( a ) and right demiphone 720 ( b ), a different level of or context dominance is accorded to the preceding phone and the succeeding phone. Therefore, right-dominant enhanced demiphone AMs 712 ( a ) may be advantageously utilized to optimally model certain phones for performing speech recognition procedures with increased accuracy characteristics. In certain embodiments, right-dominant demiphone AMs 712 ( a ) may be effectively utilized to model all supported phones that are utilized by speech recognition engine 214 . Techniques for implementing right-dominant enhanced demiphone AMs 712 ( a ) are further discussed below in conjunction with FIGS. 9-10 .
- FIG. 8 a diagram for one embodiment of a left-dominant enhanced demiphone acoustic model (AM) 712 ( b ) is shown, in accordance with one embodiment of the present invention.
- AMs 712 ( b ) may be implemented with configurations, elements, or functionalities in addition to, or instead of, certain configurations, elements, or functionalities discussed in conjunction with the FIG. 8 embodiment.
- left-dominant enhanced demiphone AM 712 ( b ) represents a given phone from a supported phone set that is used to implement acoustic models 336 and dictionary 340 from speech recognition engine 214 , as discussed above in conjunction with FIGS. 3 and 4 .
- Left-dominant enhanced demiphorie AM 712 ( b ) includes a left demiphone 720 ( c ) and a right demiphone 720 ( d ).
- the left demiphone 720 ( c ) in turn includes a first state 716 ( d ) and a second state 716 ( e ), while right demiphone 720 ( d ) includes only a third state 716 ( f ).
- First state 716 ( d ), second state 716 ( e ), and third state 716 ( f ) collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in the FIG. 8 embodiment.
- Each state 716 ( d ) and state 716 ( e ) of left demiphone AM 720 ( c ) are defined with respect to a phone context that includes information only from a preceding phone.
- state 716 ( f ) of right demiphone AM 720 ( d ) is defined with respect to a phone context that includes information from only a succeeding phone.
- left demiphone 720 ( c ) of left-dominant enhanced demiphone AM 712 ( b ) is based upon context information from only an immediately adjacent preceding phone with respect to the current phone that is modeled by left-dominant enhanced demiphone AM 712 ( b ), while right demiphone 720 ( d ) of left-dominant enhanced demiphone AM 712 ( b ) is based upon context information from only an immediately adjacent succeeding phone with respect to the current phone that is modeled by left-dominant enhanced demiphone AM 712 ( b ).
- left-dominant demiphone AMs 712 ( b ) utilize a different number of states 716 to implement the left demiphone 720 ( c ) and right demiphone 720 ( d ), a different level of or context dominance is accorded to the preceding phone and the succeeding phone. Therefore, left-dominant enhanced demiphone AMs 712 ( b ) may be advantageously utilized to optimally model certain phones for performing speech recognition procedures with increased accuracy characteristics. Techniques for implementing left-dominant enhanced demiphone AMs 712 ( b ) are further discussed below in conjunction with FIGS. 9-10 .
- decision tree 912 a block diagram for one embodiment of a decision tree 912 is shown, in accordance with the present invention.
- various types of decision trees may include other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with the FIG. 9 embodiment.
- acoustic model (AM) generator 222 may utilize decision tree 912 to determine whether a given enhanced demiphone AM 712 should be configured either as a right-dominant enhanced demiphone AM 712 ( a ) ( FIG. 7 ), or as a left-dominant enhanced demiphone AM 712 ( b ) ( FIG. 8 ).
- decision tree 912 includes a question 1 ( 916 ( a )) through a question X ( 916 ( c )).
- Decision tree 912 may be implemented to include any desired number of questions 916 that may include any desired type of information.
- a separate decision tree 912 may be created for each state of an acoustic model by analyzing linguistic characteristics and acoustic properties of the speech information in training database 226 ( FIG. 2 ) to determine either a right-dominant configuration or a left-dominant configuration for that given enhanced demiphone AM 712 within a given context.
- Each question includes a phone characteristic (such as “voiced”, “nasal”, or “consonant”), and a corresponding dominance characteristic (such as “left” or “right”).
- decision tree 912 may be implemented as a triphone decision tree with questions 916 that are based upon the states of the triphone acoustic model 512 discussed above in conjunction with FIG. 5 .
- AM generator 222 may evaluate the dominance characteristics of all questions 916 for a given decision tree 912 to determine a composite dominance characteristic for a corresponding state 716 of an enhanced demiphone AM 712 . After the dominance characteristic of each state 716 is determined, then AM generator 222 may appropriately configure the corresponding enhanced demiphone AM 712 as either a right-dominant enhanced demiphone AM 712 ( a ), or a left-dominant enhanced demiphone AM 712 ( b ). However, in certain other embodiments, decision tree 912 need not by utilized, and right-dominant demiphone AMs 712 ( a ) may be effectively utilized to model all supported phones that are utilized by speech recognition engine 214 .
- Table I illustrates dominance characteristics for three states 716 of a selected phone set of 46 phones, as determined by analyzing corresponding decision trees 912 .
- the corresponding enhanced demiphone AM 712 is configured as a right-dominant enhanced demiphone AM 712 ( a ).
- the second center state 716 is listed as “left”, then the corresponding enhanced demiphone AM 712 is configured as a left-dominant enhanced demiphone AM 712 ( b ).
- FIG. 10 a flowchart of method steps for generating enhanced demiphone acoustic models (AMs) 712 is shown, in accordance with one embodiment of the present invention.
- the FIG. 10 flowchart is presented for purposes of illustration, and in alternate embodiments, the present invention may readily utilize various steps and sequences other than certain of those discussed in conjunction with the FIG. 10 embodiment.
- a phone set of specific phones are selectively defined for utilization by a speech recognition engine 214 to perform various speech recognition procedures.
- a representative training database 226 ( FIG. 2 ) is created or provided by utilizing any effective techniques.
- an acoustic model generator 222 FIG. 2 or other appropriate entity analyzes training database 226 to generate a separate decision tree 912 for each state of acoustic models that represent the defined phone set.
- acoustic model generator 222 analyzes decision trees 912 to define appropriate dominance configurations for corresponding enhanced demiphone acoustic models 712 .
- the foregoing configurations include a right-dominant enhanced demiphone configuration (see FIG. 7 ), and a left-dominant enhanced demiphone configuration (see FIG. 8 ).
- acoustic model generator 222 implements the enhanced demiphone acoustic models 712 using the dominance configurations previously defined in step 1022 .
- speech recognition engine 214 may then utilize the enhanced demiphone acoustic models 712 to accurately and efficiently perform speech recognition procedures.
- the FIG. 10 process may then terminate.
- the present invention therefore provides an improved system and method for generating enhanced demiphone acoustic models for speech recognition.
Abstract
Description
- 1. Field of Invention
- This invention relates generally to electronic speech recognition systems, and relates more particularly to a methodology for generating enhanced demiphone acoustic models for speech recognition.
- 2. Background
- Implementing robust and effective techniques for system users to interface with electronic devices is a significant consideration of system designers and manufacturers. Voice-controlled operation of electronic devices often provides a desirable interface for system users to control and interact with electronic devices. For example, voice-controlled operation of an electronic device may allow a user to perform other tasks simultaneously, or can be advantageous in certain types of operating environments. In addition, hands-free operation of electronic devices may also be desirable for users who have physical limitations or other special requirements.
- Hands-free operation of electronic devices may be implemented by various speech-activated electronic devices. Speech-activated electronic devices advantageously allow users to interface with electronic devices in situations where it would be inconvenient or potentially hazardous to utilize a traditional input device. However, effectively implementing such speech recognition systems creates substantial challenges for system designers.
- For example, enhanced demands for increased system functionality and performance require more system processing power and require additional memory resources. An increase in processing or memory requirements typically results in a corresponding detrimental economic impact due to increased production costs and operational inefficiencies.
- Furthermore, enhanced system capability to perform various advanced operations provides additional benefits to a system user, but may also place increased demands on the control and management of various system components. Therefore, for at least the foregoing reasons, implementing a robust and effective method for a system user to interface with electronic devices through speech recognition remains a significant consideration of system designers and manufacturers.
- In accordance with the present invention, a methodology is disclosed for generating enhanced demiphone acoustic models for speech recognition. In one embodiment, a phone set of specific phones are selectively defined for utilization by a speech recognition engine to perform various speech recognition procedures. In addition, a representative training database is created or provided by utilizing any effective techniques. Then, an acoustic model generator or other appropriate entity analyzes the training database to generate a separate decision tree for each state of acoustic models that represent the defined phone set.
- Next, the acoustic model generator analyzes the decision trees to define appropriate dominance configurations for corresponding enhanced demiphone acoustic models. In certain embodiments, the foregoing configurations include a preceding-dominant enhanced demiphone configuration and a succeeding-dominant enhanced demiphone configuration. The acoustic model generator then implements the enhanced demiphone acoustic models using the dominance configurations previously defined.
- In certain embodiments, the succeeding-dominant (right-dominant) enhanced demiphone includes a preceding (left) demiphone with relatively fewer states (e.g. one or zero states) and a succeeding (right) demiphone with relatively more states (e.g. two or three states). In addition, the preceding-dominant (left-dominant) enhanced demiphone includes a preceding (left) demiphone with relatively more states (e.g. two or three states) and a succeeding (right) demiphone with relatively fewer states (zero or one state). The speech recognition engine may then utilize the enhanced demiphone acoustic models to accurately and efficiently perform speech recognition procedures. For at least the foregoing reasons, the present invention therefore provides an improved system and method for generating enhanced demiphone acoustic models for speech recognition.
-
FIG. 1 is a block diagram for one embodiment of an electronic device, in accordance with the present invention; -
FIG. 2 is a block diagram for one embodiment of the memory ofFIG. 1 , in accordance with the present invention; -
FIG. 3 is a block diagram for one embodiment of the speech recognition engine ofFIG. 2 , in accordance with the present invention; -
FIG. 4 is a block diagram illustrating functionality of the speech recognition engine ofFIG. 3 , in accordance with one embodiment of the present invention; -
FIG. 5 is a diagram for one embodiment of a triphone acoustic model; -
FIG. 6 is a diagram for one embodiment of a conventional demiphone acoustic model; -
FIG. 7 is a diagram for one embodiment of a right-dominant enhanced demiphone acoustic model, in accordance with the present invention; -
FIG. 8 is a diagram for one embodiment of a left-dominant enhanced demiphone acoustic model, in accordance with the present invention; -
FIG. 9 is a diagram for one embodiment of a decision tree, in accordance with the present invention; and -
FIG. 10 is a flowchart of method steps for generating enhanced demiphone acoustic models for speech recognition, in accordance with one embodiment of the present invention. - The present invention relates to an improvement in speech recognition systems. The following description is presented to enable one of ordinary skill in the art to make and use the invention, and is provided in the context of a patent application and its requirements. Various modifications to the embodiments disclosed herein will be apparent to those skilled in the art, and the generic principles herein may be applied to other embodiments. Thus, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features described herein.
- The present invention comprises a system and method for effectively performing speech recognition procedures, and includes enhanced demiphone acoustic models that a speech recognition engine utilizes to perform the speech recognition procedures. The enhanced demiphone acoustic models each have three states that are collectively arranged to form a left demiphone and a right demiphone. An acoustic model generator utilizes a decision tree for analyzing speech context information from a training database. To accurately model speech characteristics, the acoustic model generator then configures each of the enhanced demiphone acoustic models as either a right-dominant enhanced demiphone acoustic model or a left-dominant enhanced demiphone acoustic model.
- Referring now to
FIG. 1 , a block diagram for one embodiment of anelectronic device 110 is shown, according to the present invention. TheFIG. 1 embodiment includes, but is not limited to, asound sensor 112, acontrol module 114, and adisplay 134. In alternate embodiments,electronic device 110 may readily include various other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with theFIG. 1 embodiment. - In accordance with certain embodiments of the present invention,
electronic device 110 may be embodied as any appropriate electronic device or system. For example, in certain embodiments,electronic device 110 may be implemented as a computer device, a personal digital assistant (PDA), a cellular telephone, a television, a game console, and as part of entertainment robots such as AIBO™ and QRIO™ by Sony Corporation. - In the
FIG. 1 embodiment,electronic device 110 utilizessound sensor 112 to detect and convert ambient sound energy into corresponding audio data. The captured audio data is then transferred oversystem bus 124 toCPU 122, which responsively performs various processes and functions with the captured audio data, in accordance with the present invention. - In the
FIG. 1 embodiment,control module 114 includes, but is not limited to, a central processing unit (CPU) 122 (electronic data processor), amemory 130, and one or more input/output interface(s) (I/O) 126.Display 134,CPU 122,memory 130, and I/O 126 are each coupled to, and communicate, viacommon system bus 124. In alternate embodiments,control module 114 may readily include various other components in addition to, or instead of, those components discussed in conjunction with theFIG. 1 embodiment. - In the
FIG. 1 embodiment,CPU 122 is implemented to include any appropriate microprocessor device. Alternately,CPU 122 may be implemented using any other appropriate technology. For example,CPU 122 may be implemented as an application-specific integrated circuit (ASIC) or other appropriate electronic device. In theFIG. 1 embodiment, I/O 126 provides one or more effective interfaces for facilitating bi-directional communications betweenelectronic device 110 and any external entity, including a system user or another electronic device. I/O 126 may be implemented using any appropriate input and/or output devices. The functionality and utilization ofelectronic device 110 are further discussed below in conjunction withFIGS. 2-10 . - Referring now to
FIG. 2 , a block diagram for one embodiment of theFIG. 1 memory 130 is shown, according to the present invention.Memory 130 may comprise any desired storage-device configurations, including, but not limited to, random access memory (RAM), read-only memory (ROM), and storage devices such as floppy discs or hard disc drives, and SONY® MEMORY STICK®. In theFIG. 2 embodiment,memory 130 stores adevice application 210, aspeech recognition engine 214, an acoustic model (AM)generator 222, and atraining database 226. In alternate embodiments,memory 130 may readily store other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with theFIG. 2 embodiment. - In the
FIG. 2 embodiment,device application 210 includes program instructions that are executed by CPU 122 (FIG. 1 ) to perform various functions and operations forelectronic device 110. The particular nature and functionality ofdevice application 210 typically varies depending upon factors such as the type and particular use of the correspondingelectronic device 110. - In the
FIG. 2 embodiment,speech recognition engine 214 includes one or more software modules that are executed byCPU 122 to analyze and recognize input sound data. Certain embodiments ofspeech recognition engine 214 are further discussed below in conjunction withFIGS. 3-4 . In theFIG. 2 embodiment,electronic device 110 may utilizeAM generator 222 to generate enhanced demiphone acoustic models for use byspeech recognition engine 214 in performing speech recognition procedures. In theFIG. 2 embodiment,training database 226 includes an extensive selection of representative recorded speech samples. The generation of enhanced demiphone acoustic models byAM generator 222 is further discussed below in conjunction withFIGS. 7-10 . - Referring now to
FIG. 3 , a block diagram for one embodiment of theFIG. 2 speech recognition engine 214 is shown, in accordance with the present invention.Speech recognition engine 214 includes, but is not limited to, afeature extractor 310, anendpoint detector 312, arecognizer 314,acoustic models 336,dictionary 340, andlanguage models 344. In alternate embodiments,speech recognition engine 214 may readily include various other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with theFIG. 3 embodiment. - In the
FIG. 3 embodiment, sound sensor 112 (FIG. 1 ) provides digital speech data to featureextractor 310 viasystem bus 124.Feature extractor 310 responsively generates corresponding representative feature vectors, which are provided to recognizer 314 viapath 320.Feature extractor 310 further provides the speech data toendpoint detector 312, andendpoint detector 312 responsively identifies endpoints of utterances represented by the speech data to indicate the beginning and end of an utterance in time.Endpoint detector 312 then provides the endpoints torecognizer 314. - In the
FIG. 3 embodiment,recognizer 314 is configured to recognize words in a vocabulary that is represented indictionary 340. The vocabulary represented indictionary 340 corresponds to any desired sentences, word sequences, commands, instructions, narration, or other audible sounds that are supported for speech recognition byspeech recognition engine 214. - In practice, each word from
dictionary 340 is associated with a corresponding phone string (string of individual phones) which represents the pronunciation of that word. Acoustic models 336 (such as Hidden Markov Models) for each of the phones are selected and combined to create the foregoing phone strings for accurately representing pronunciations of words indictionary 340. In accordance with the present invention,acoustic models 336 are implemented as enhanced demiphone acoustic models.Recognizer 314 compares input feature vectors fromline 320 with the entries (phone strings) fromdictionary 340 to determine which word produces the highest recognition score. The word corresponding to the highest recognition score may thus be identified as the recognized word. -
Speech recognition engine 214 also utilizeslanguage models 344 as a recognition grammar to determine specific recognized word sequences that are supported byspeech recognition engine 214. The recognized sequences of vocabulary words may then be output as recognition results fromrecognizer 314 viapath 332. The operation and implementation ofrecognizer 314,dictionary 340, andlanguage models 344 are further discussed below in conjunction withFIG. 4 . - Referring now to
FIG. 4 , a block diagram illustrating functionality of theFIG. 3 speech recognition engine 214 is shown, in accordance with one embodiment of the present invention. In alternate embodiments, the present invention may readily perform speech recognition procedures using various techniques or functionalities in addition to, or instead of, certain techniques or functionalities discussed in conjunction with theFIG. 4 embodiment. - In the
FIG. 4 embodiment,speech recognition engine 214 receives speech data fromsound sensor 112, as discussed above in conjunction withFIG. 3 . Recognizer 314 (FIG. 3 ) fromspeech recognition engine 214 sequentially compares segments of the input speech data withacoustic models 336 to identify a series of phones (phone strings) that represent the input speech data. In accordance with the present invention,acoustic models 336 are implemented as enhanced demiphone acoustic models, as discussed below in conjunction withFIGS. 7-10 . -
Recognizer 314references dictionary 340 to look up recognized vocabulary words that correspond to the identified phone strings. Therecognizer 314 then utilizeslanguage models 344 as a recognition grammar to form the recognized vocabulary words into word sequences, such as sentences, phrases, commands, or narration, which are supported byspeech recognition engine 214. Various techniques for effectively generating enhanced demiphone acoustic models are further discussed below in conjunction withFIGS. 7-10 . - Referring now to
FIG. 5 , a diagram for one embodiment of a triphone acoustic model (AM) 512 is shown. In theFIG. 5 embodiment,triphone AM 512 represents a given phone from a supported phone set that is used to implement a speech recognition engine.Triphone AM 512 includes a first state 516(a), a second state 516(b), and a third state 516(c) that collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in theFIG. 5 embodiment. - Each
state 516 oftriphone AM 512 is defined with respect to a phone context that includes information from both a preceding phone and a succeeding phone. In other words, all threestates 516 of the current phone that is modeled bytriphone AM 512 are based upon context information from both an immediately preceding phone and an immediately succeeding phone. - Because
triphone AM 512 is implemented with three discrete states, in order to implement a phone set with a given number of phones “X”, the total number oftriphone AMs 512 required by a corresponding speech recognition engine is expressed by the following formula: Total Triphone AMs=X3. Utilizingtriphone AMs 512 for performing speech recognition procedures may therefore require a relatively large amount of memory and processing resources. - Referring now to
FIG. 6 , a diagram for one embodiment of a conventional demiphone acoustic model (AM) 612 is shown. In theFIG. 6 embodiment,conventional demiphone AM 612 represents a given phone from a supported phone set that is used to implement a speech recognition engine.Conventional demiphone AM 612 includes a conventional left demiphone 620(a) and a conventional right demiphone 620(b). The conventional left demiphone 620(a) in turn includes a first state 616(a) and a second state 616(b), while conventional right demiphone 620(b) includes a third state 616(c) and a fourth state 616(d). First state 616(a), second state 616(b), third state 616(c), and fourth state 616(d) collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in theFIG. 6 embodiment. - Each state 616(a) and 616(b) of conventional left demiphone AM 620(a) is defined with respect to a phone context that includes information only from a preceding phone. Similarly, each state 616(c) and 616(d) of conventional right demiphone AM 620(b) is defined with respect to a phone context that includes information from only a succeeding phone. In other words, left demiphone 620(a) of
conventional demiphone AM 612 is based only upon context information from an immediately preceding phone with respect to the current phone that is modeled byconventional demiphone AM 612, and right demiphone 620(b) ofconventional demiphone AM 612 is based only upon context information from an immediately succeeding phone with respect to the current phone that is modeled byconventional demiphone AM 612. - Because
conventional demiphone AM 612 is implemented with twodiscrete demiphones 620, in order to implement a phone set with a given number of phones “X”, the total number ofconventional demiphone AMs 612 required by a corresponding speech recognition engine is expressed by the following formula: Total Demiphone AMs=2*(X2). Utilizingconventional demiphone AMs 612 for performing speech recognition procedures may therefore conserve memory and processing resources as compared to a speech recognition system that utilizes triphone AMs 512 (FIG. 5 ). - However, because
conventional demiphone AMs 612 utilize the same number ofstates 616 to implement both conventional left demiphone 620(a) and conventional right demiphone 620(b), an equal level of dominance is accorded to both the preceding phone and the succeeding phone. In actuality, during most speech contexts, the foregoing equal level of dominance does not accurately reflect true acoustic characteristics of speech. Therefore,conventional demiphone AMs 612 do not optimally model speech, and speech recognition procedures may exhibit reduced accuracy as a result. To address this problem, improved techniques for implementing enhanced demiphone acoustic models are discussed below in conjunction withFIGS. 7-10 . - Referring now to
FIG. 7 , a diagram for one embodiment of a right-dominant enhanced demiphone acoustic model (AM) 712(a) is shown, in accordance with one embodiment of the present invention. In alternate embodiments, various right-dominant enhanced demiphone AMs 712(a) may be implemented with configurations, elements, or functionalities in addition to, or instead of, certain configurations, elements, or functionalities discussed in conjunction with theFIG. 7 embodiment. - In the
FIG. 7 embodiment, right-dominant enhanced demiphone AM 712(a) represents a given phone from a supported phone set that is used to implementacoustic models 336 anddictionary 340 fromspeech recognition engine 214, as discussed above in conjunction withFIGS. 3 and 4 . Right-dominant enhanced demiphone AM 712(a) includes a left demiphone 720(a) and a right demiphone 720(b). The left demiphone 720(a) in turn includes only a first state 716(a), while right demiphone 720(b) includes a second state 716(b) and a third state 716(c). First state 716(a), second state 716(b), and third state 716(c) collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in theFIG. 7 embodiment. - State 716(a) of left demiphone AM 720(a) is defined with respect to a phone context that includes information only from a preceding phone. Similarly, each state 716(b) and 716(c) of right demiphone AM 720(b) is defined with respect to a phone context that includes information from only a succeeding phone. In other words, left demiphone 720(a) of right-dominant enhanced demiphone AM 712(a) is based upon context information from only an immediately preceding phone with respect to the current phone that is modeled by right-dominant enhanced demiphone AM 712(a), and right demiphone 720(b) of right-dominant enhanced demiphone AM 712(a) is based upon context information from only an immediately succeeding phone with respect to the current phone that is modeled by right-dominant enhanced demiphone AM 712(a).
- Because right-dominant enhanced demiphone AM 712(a) is implemented with two
discrete demiphones 720, in order to implement a phone set with a given number of phones “X”, the total number of enhanceddemiphone AMs 712 required by a correspondingspeech recognition engine 214 is expressed by the following formula: Total Demiphone AMs=2*(X2). Utilizing enhanced demiphone AMs 712(a) for performing speech recognition procedures may therefore conserve memory and processing resources as compared to a speech recognition system that utilizes triphone AMs 512 (FIG. 5 ). - In addition, because right-dominant demiphone AMs 712(a) utilize a different number of
states 716 to implement left demiphone 720(a) and right demiphone 720(b), a different level of or context dominance is accorded to the preceding phone and the succeeding phone. Therefore, right-dominant enhanced demiphone AMs 712(a) may be advantageously utilized to optimally model certain phones for performing speech recognition procedures with increased accuracy characteristics. In certain embodiments, right-dominant demiphone AMs 712(a) may be effectively utilized to model all supported phones that are utilized byspeech recognition engine 214. Techniques for implementing right-dominant enhanced demiphone AMs 712(a) are further discussed below in conjunction withFIGS. 9-10 . - Referring now to
FIG. 8 , a diagram for one embodiment of a left-dominant enhanced demiphone acoustic model (AM) 712(b) is shown, in accordance with one embodiment of the present invention. In alternate embodiments, various left-dominant enhanced demiphone AMs 712(b) may be implemented with configurations, elements, or functionalities in addition to, or instead of, certain configurations, elements, or functionalities discussed in conjunction with theFIG. 8 embodiment. - In the
FIG. 8 embodiment, left-dominant enhanced demiphone AM 712(b) represents a given phone from a supported phone set that is used to implementacoustic models 336 anddictionary 340 fromspeech recognition engine 214, as discussed above in conjunction withFIGS. 3 and 4 . Left-dominant enhanced demiphorie AM 712(b) includes a left demiphone 720(c) and a right demiphone 720(d). The left demiphone 720(c) in turn includes a first state 716(d) and a second state 716(e), while right demiphone 720(d) includes only a third state 716(f). First state 716(d), second state 716(e), and third state 716(f) collectively model the corresponding phone in a temporal sequence that progresses from left to right as depicted in theFIG. 8 embodiment. - Each state 716(d) and state 716(e) of left demiphone AM 720(c) are defined with respect to a phone context that includes information only from a preceding phone. Similarly, state 716(f) of right demiphone AM 720(d) is defined with respect to a phone context that includes information from only a succeeding phone. In other words, left demiphone 720(c) of left-dominant enhanced demiphone AM 712(b) is based upon context information from only an immediately adjacent preceding phone with respect to the current phone that is modeled by left-dominant enhanced demiphone AM 712(b), while right demiphone 720(d) of left-dominant enhanced demiphone AM 712(b) is based upon context information from only an immediately adjacent succeeding phone with respect to the current phone that is modeled by left-dominant enhanced demiphone AM 712(b).
- Because left-dominant enhanced demiphone AM 712(b) is implemented with two
discrete demiphones 720, in order to implement a phone set with a given number of phones “X”, the total number of enhanceddemiphone AMs 712 required by a correspondingspeech recognition engine 214 may be expressed by the following formula: Total Demiphone AMs=2*(X2). Utilizing enhanced demiphone AMs 712(b) for performing speech recognition procedures may therefore conserve memory and processing resources when compared to a speech recognition system that utilizes triphone AMs 512 (FIG. 5 ). - In addition, because left-dominant demiphone AMs 712(b) utilize a different number of
states 716 to implement the left demiphone 720(c) and right demiphone 720(d), a different level of or context dominance is accorded to the preceding phone and the succeeding phone. Therefore, left-dominant enhanced demiphone AMs 712(b) may be advantageously utilized to optimally model certain phones for performing speech recognition procedures with increased accuracy characteristics. Techniques for implementing left-dominant enhanced demiphone AMs 712(b) are further discussed below in conjunction withFIGS. 9-10 . - Referring now to
FIG. 9 , a block diagram for one embodiment of adecision tree 912 is shown, in accordance with the present invention. In alternate embodiments, various types of decision trees may include other elements or functionalities in addition to, or instead of, certain elements or functionalities discussed in conjunction with theFIG. 9 embodiment. - In accordance with certain embodiments of the present invention, acoustic model (AM) generator 222 (
FIG. 2 ) may utilizedecision tree 912 to determine whether a givenenhanced demiphone AM 712 should be configured either as a right-dominant enhanced demiphone AM 712(a) (FIG. 7 ), or as a left-dominant enhanced demiphone AM 712(b) (FIG. 8 ). In theFIG. 9 embodiment,decision tree 912 includes a question 1 (916(a)) through a question X (916(c)).Decision tree 912 may be implemented to include any desired number ofquestions 916 that may include any desired type of information. - In the
FIG. 9 embodiment, aseparate decision tree 912 may be created for each state of an acoustic model by analyzing linguistic characteristics and acoustic properties of the speech information in training database 226 (FIG. 2 ) to determine either a right-dominant configuration or a left-dominant configuration for that givenenhanced demiphone AM 712 within a given context. Each question includes a phone characteristic (such as “voiced”, “nasal”, or “consonant”), and a corresponding dominance characteristic (such as “left” or “right”). - A left dominance indicates primary influence from a preceding phone with respect to the current acoustic model, and a right dominance indicates primary influence from a succeeding phone with respect to the current acoustic model. In certain embodiments,
decision tree 912 may be implemented as a triphone decision tree withquestions 916 that are based upon the states of the triphoneacoustic model 512 discussed above in conjunction withFIG. 5 . - In accordance with the present invention, AM generator 222 (
FIG. 2 ) or other appropriate entity may evaluate the dominance characteristics of allquestions 916 for a givendecision tree 912 to determine a composite dominance characteristic for acorresponding state 716 of anenhanced demiphone AM 712. After the dominance characteristic of eachstate 716 is determined, thenAM generator 222 may appropriately configure the correspondingenhanced demiphone AM 712 as either a right-dominant enhanced demiphone AM 712(a), or a left-dominant enhanced demiphone AM 712(b). However, in certain other embodiments,decision tree 912 need not by utilized, and right-dominant demiphone AMs 712(a) may be effectively utilized to model all supported phones that are utilized byspeech recognition engine 214. - The following Table I illustrates dominance characteristics for three
states 716 of a selected phone set of 46 phones, as determined by analyzing correspondingdecision trees 912. In accordance with the present invention, if thesecond center state 716 is listed as “right”, then the correspondingenhanced demiphone AM 712 is configured as a right-dominant enhanced demiphone AM 712(a). Similarly, if thesecond center state 716 is listed as “left”, then the correspondingenhanced demiphone AM 712 is configured as a left-dominant enhanced demiphone AM 712(b).TABLE I Phone 1st state 2nd state 3rd state aa left right right ae left right right ah left left right ao left right right aw left left right ax left right right ay left left right b left right right ch left right right d left right right dh left left right eh left right right er left right right ey left right right f left right right g left right right hh left right right ih left left right ix left right right iy left right right jh left right right k left right right l left left right m left right right n left right right ng left right right ow left right right oy left monophone right p left right right r left left right s left right right sh left left right t left right right th left left right uh left left left uw left right right v left right right w left right right y left left right Z left right right Zh monophone monophone monophone cl (closure) left left right - Referring now to
FIG. 10 , a flowchart of method steps for generating enhanced demiphone acoustic models (AMs) 712 is shown, in accordance with one embodiment of the present invention. TheFIG. 10 flowchart is presented for purposes of illustration, and in alternate embodiments, the present invention may readily utilize various steps and sequences other than certain of those discussed in conjunction with theFIG. 10 embodiment. - In the
FIG. 10 embodiment, instep 1010, a phone set of specific phones are selectively defined for utilization by aspeech recognition engine 214 to perform various speech recognition procedures. In addition, instep 1014, a representative training database 226 (FIG. 2 ) is created or provided by utilizing any effective techniques. Instep 1018, an acoustic model generator 222 (FIG. 2 ) or other appropriate entity analyzestraining database 226 to generate aseparate decision tree 912 for each state of acoustic models that represent the defined phone set. - In
step 1022,acoustic model generator 222 analyzesdecision trees 912 to define appropriate dominance configurations for corresponding enhanced demiphoneacoustic models 712. In certain embodiments, the foregoing configurations include a right-dominant enhanced demiphone configuration (seeFIG. 7 ), and a left-dominant enhanced demiphone configuration (seeFIG. 8 ). Next, instep 1026,acoustic model generator 222 implements the enhanced demiphoneacoustic models 712 using the dominance configurations previously defined instep 1022. Finally, instep 1030,speech recognition engine 214 may then utilize the enhanced demiphoneacoustic models 712 to accurately and efficiently perform speech recognition procedures. TheFIG. 10 process may then terminate. The present invention therefore provides an improved system and method for generating enhanced demiphone acoustic models for speech recognition. - The invention has been explained above with reference to certain preferred embodiments. Other embodiments will be apparent to those skilled in the art in light of this disclosure. For example, the present invention may readily be implemented using configurations and techniques other than those described in the embodiments above. Additionally, the present invention may effectively be used in conjunction with systems other than those described above as the preferred embodiments. Therefore, these and other variations upon the foregoing embodiments are intended to be covered by the present invention, which is limited only by the appended claims.
Claims (43)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/013,888 US7467086B2 (en) | 2004-12-16 | 2004-12-16 | Methodology for generating enhanced demiphone acoustic models for speech recognition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/013,888 US7467086B2 (en) | 2004-12-16 | 2004-12-16 | Methodology for generating enhanced demiphone acoustic models for speech recognition |
Publications (2)
Publication Number | Publication Date |
---|---|
US20060136209A1 true US20060136209A1 (en) | 2006-06-22 |
US7467086B2 US7467086B2 (en) | 2008-12-16 |
Family
ID=36597231
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/013,888 Expired - Fee Related US7467086B2 (en) | 2004-12-16 | 2004-12-16 | Methodology for generating enhanced demiphone acoustic models for speech recognition |
Country Status (1)
Country | Link |
---|---|
US (1) | US7467086B2 (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150127349A1 (en) * | 2013-11-01 | 2015-05-07 | Google Inc. | Method and System for Cross-Lingual Voice Conversion |
US20150127350A1 (en) * | 2013-11-01 | 2015-05-07 | Google Inc. | Method and System for Non-Parametric Voice Conversion |
EP2916530A1 (en) * | 2014-03-05 | 2015-09-09 | ASC telecom AG | Method for controlling the recording of communication data |
US9240184B1 (en) | 2012-11-15 | 2016-01-19 | Google Inc. | Frame-level combination of deep neural network and gaussian mixture models |
US9542927B2 (en) | 2014-11-13 | 2017-01-10 | Google Inc. | Method and system for building text-to-speech voice from diverse recordings |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7664643B2 (en) * | 2006-08-25 | 2010-02-16 | International Business Machines Corporation | System and method for speech separation and multi-talker speech recognition |
JP4427530B2 (en) * | 2006-09-21 | 2010-03-10 | 株式会社東芝 | Speech recognition apparatus, program, and speech recognition method |
US8719023B2 (en) | 2010-05-21 | 2014-05-06 | Sony Computer Entertainment Inc. | Robustness to environmental changes of a context dependent speech recognizer |
US8484022B1 (en) | 2012-07-27 | 2013-07-09 | Google Inc. | Adaptive auto-encoders |
US8442821B1 (en) | 2012-07-27 | 2013-05-14 | Google Inc. | Multi-frame prediction for hybrid neural network/hidden Markov models |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5345537A (en) * | 1990-12-19 | 1994-09-06 | Fujitsu Limited | Network reformer and creator |
US5440663A (en) * | 1992-09-28 | 1995-08-08 | International Business Machines Corporation | Computer system for speech recognition |
US5459809A (en) * | 1993-03-15 | 1995-10-17 | Jin-Hyung Kim | Character recognition system and method therefor accommodating on-line discrete and cursive handwritten |
US6151574A (en) * | 1997-12-05 | 2000-11-21 | Lucent Technologies Inc. | Technique for adaptation of hidden markov models for speech recognition |
US6285981B1 (en) * | 1998-06-30 | 2001-09-04 | Texas Instruments Incorporated | Speed up speech recognition search using macro evaluator |
US20020046017A1 (en) * | 2000-04-03 | 2002-04-18 | Xerox Corporation | Method and apparatus for aligning ambiguity in finite state transducers |
US6665641B1 (en) * | 1998-11-13 | 2003-12-16 | Scansoft, Inc. | Speech synthesis using concatenation of speech waveforms |
US20040172249A1 (en) * | 2001-05-25 | 2004-09-02 | Taylor Paul Alexander | Speech synthesis |
US7089183B2 (en) * | 2000-08-02 | 2006-08-08 | Texas Instruments Incorporated | Accumulating transformations for hierarchical linear regression HMM adaptation |
US7275034B2 (en) * | 2003-02-11 | 2007-09-25 | Microsoft Corporation | Word-specific acoustic models in a speech recognition system |
-
2004
- 2004-12-16 US US11/013,888 patent/US7467086B2/en not_active Expired - Fee Related
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5345537A (en) * | 1990-12-19 | 1994-09-06 | Fujitsu Limited | Network reformer and creator |
US5440663A (en) * | 1992-09-28 | 1995-08-08 | International Business Machines Corporation | Computer system for speech recognition |
US5459809A (en) * | 1993-03-15 | 1995-10-17 | Jin-Hyung Kim | Character recognition system and method therefor accommodating on-line discrete and cursive handwritten |
US6151574A (en) * | 1997-12-05 | 2000-11-21 | Lucent Technologies Inc. | Technique for adaptation of hidden markov models for speech recognition |
US6285981B1 (en) * | 1998-06-30 | 2001-09-04 | Texas Instruments Incorporated | Speed up speech recognition search using macro evaluator |
US6665641B1 (en) * | 1998-11-13 | 2003-12-16 | Scansoft, Inc. | Speech synthesis using concatenation of speech waveforms |
US20040111266A1 (en) * | 1998-11-13 | 2004-06-10 | Geert Coorman | Speech synthesis using concatenation of speech waveforms |
US20020046017A1 (en) * | 2000-04-03 | 2002-04-18 | Xerox Corporation | Method and apparatus for aligning ambiguity in finite state transducers |
US7089183B2 (en) * | 2000-08-02 | 2006-08-08 | Texas Instruments Incorporated | Accumulating transformations for hierarchical linear regression HMM adaptation |
US20040172249A1 (en) * | 2001-05-25 | 2004-09-02 | Taylor Paul Alexander | Speech synthesis |
US7275034B2 (en) * | 2003-02-11 | 2007-09-25 | Microsoft Corporation | Word-specific acoustic models in a speech recognition system |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9240184B1 (en) | 2012-11-15 | 2016-01-19 | Google Inc. | Frame-level combination of deep neural network and gaussian mixture models |
US20150127349A1 (en) * | 2013-11-01 | 2015-05-07 | Google Inc. | Method and System for Cross-Lingual Voice Conversion |
US20150127350A1 (en) * | 2013-11-01 | 2015-05-07 | Google Inc. | Method and System for Non-Parametric Voice Conversion |
US9177549B2 (en) * | 2013-11-01 | 2015-11-03 | Google Inc. | Method and system for cross-lingual voice conversion |
US9183830B2 (en) * | 2013-11-01 | 2015-11-10 | Google Inc. | Method and system for non-parametric voice conversion |
EP2916530A1 (en) * | 2014-03-05 | 2015-09-09 | ASC telecom AG | Method for controlling the recording of communication data |
US9542927B2 (en) | 2014-11-13 | 2017-01-10 | Google Inc. | Method and system for building text-to-speech voice from diverse recordings |
Also Published As
Publication number | Publication date |
---|---|
US7467086B2 (en) | 2008-12-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7392186B2 (en) | System and method for effectively implementing an optimized language model for speech recognition | |
US7043431B2 (en) | Multilingual speech recognition system using text derived recognition models | |
US7869999B2 (en) | Systems and methods for selecting from multiple phonectic transcriptions for text-to-speech synthesis | |
JP5174937B2 (en) | Speech recognition apparatus and method | |
JP2002304190A (en) | Method for generating pronunciation change form and method for speech recognition | |
JP2003022087A (en) | Voice recognition method | |
US7181395B1 (en) | Methods and apparatus for automatic generation of multiple pronunciations from acoustic data | |
Darjaa et al. | Effective triphone mapping for acoustic modeling in speech recognition | |
JP2001521193A (en) | Parameter sharing speech recognition method and apparatus | |
WO2007005098A2 (en) | Method and apparatus for generating and updating a voice tag | |
JP2004198831A (en) | Method, program, and recording medium for speech recognition | |
US7467086B2 (en) | Methodology for generating enhanced demiphone acoustic models for speech recognition | |
US6631348B1 (en) | Dynamic speech recognition pattern switching for enhanced speech recognition accuracy | |
JP4499389B2 (en) | Method and apparatus for generating decision tree questions for speech processing | |
US20040193416A1 (en) | System and method for speech recognition utilizing a merged dictionary | |
KR100930714B1 (en) | Voice recognition device and method | |
Viikki et al. | Speaker-and language-independent speech recognition in mobile communication systems | |
US7353173B2 (en) | System and method for Mandarin Chinese speech recognition using an optimized phone set | |
US7272560B2 (en) | Methodology for performing a refinement procedure to implement a speech recognition dictionary | |
JP2007078943A (en) | Acoustic score calculating program | |
US20060136210A1 (en) | System and method for tying variance vectors for speech recognition | |
US8099280B2 (en) | Speech recognition method and speech recognition apparatus | |
EP3718107B1 (en) | Speech signal processing and evaluation | |
US20060031071A1 (en) | System and method for automatically implementing a finite state automaton for speech recognition | |
JP2005091758A (en) | System and method for speaker recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AS | Assignment |
Owner name: SONY CORPORATION, JAPAN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:XAVIER MENENDEZ-PIDAL;OLORENSHAW, LEX S.;ABREGO, GUSTAVO HERNANDEZ;REEL/FRAME:016104/0761;SIGNING DATES FROM 20041213 TO 20041214 Owner name: SONY ELECTRONICS INC., NEW JERSEY Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:XAVIER MENENDEZ-PIDAL;OLORENSHAW, LEX S.;ABREGO, GUSTAVO HERNANDEZ;REEL/FRAME:016104/0761;SIGNING DATES FROM 20041213 TO 20041214 |
|
FEPP | Fee payment procedure |
Free format text: PAYOR NUMBER ASSIGNED (ORIGINAL EVENT CODE: ASPN); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY |
|
REMI | Maintenance fee reminder mailed | ||
LAPS | Lapse for failure to pay maintenance fees | ||
STCH | Information on status: patent discontinuation |
Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362 |
|
FP | Lapsed due to failure to pay maintenance fee |
Effective date: 20121216 |